Ask, Don't Judge: Binary Questions for Interpretable LLM Evaluation and Self-Improvement

2026-06-25

AI Summary

This work addresses key challenges in large language model (LLM) evaluation—namely high costs, low correlation between automated metrics and human judgments, and opaque scoring—by introducing BINEVAL, a novel framework that decomposes evaluation criteria into atomic binary questions. These questions are answered independently by an LLM and then aggregated via a calibration mechanism to yield multidimensional, interpretable scores. Requiring no training and agnostic to specific tasks, BINEVAL enables diagnostic analysis and prompt self-optimization while circumventing the ceiling effects inherent in conventional scoring approaches. Evaluated on benchmarks such as SummEval, Topical-Chat, and QAGS, BINEVAL matches or surpasses strong baselines like UniEval and G-Eval, demonstrating superior alignment with human ratings—particularly in factual consistency—and significantly improved discrimination between borderline and clearly erroneous outputs.

Abstract

Evaluating LLM outputs remains a major bottleneck in NLP: human evaluation is expensive and slow, lexical metrics correlate poorly with human judgments on open-ended generation, and holistic LLM judges often produce opaque scores that are hard to debug. We propose BINEVAL, a framework that decomposes evaluation criteria into atomic binary questions and aggregates the resulting verdicts into interpretable, multi-dimensional scores. Given a task prompt, a meta-prompt generates fine-grained evaluation questions, and an LLM answers them independently for each output, yielding transparent question-level feedback together with calibrated overall scores. This decomposition makes evaluation easier to inspect, easier to diagnose, and directly usable for prompt improvement. Across SummEval, Topical-Chat, and QAGS, BINEVAL matches or outperforms strong baselines including UniEval and G-Eval, with especially strong results on factual consistency benchmarks such as QAGS. Beyond competitive correlation with human judgments, BINEVAL better matches human score distributions and avoids the ceiling effects common in prior LLM judges, leading to better discrimination between borderline and clearly flawed outputs. We further show that the same question-level feedback supports iterative prompt optimization, improving evaluator prompts on summarization and generation prompts on IFBench under both self-update and cross-model update settings. Overall, BINEVAL provides a task-agnostic, training-free, and interpretable evaluation framework that combines strong empirical performance with practical diagnostic and optimization value.

Problem

• LLM evaluation
• interpretable evaluation
• binary questions
• factual consistency
• evaluation bottleneck

Innovation

• binary questions
• interpretable evaluation
• LLM self-improvement
• prompt optimization
• factual consistency